In distributed management systems, where a central management server initiates management tasks on a set of endpoints, it is common to have tasks that over-subscribe the network link of the management server when run in parallel against all endpoints. System managers value both throughput (end-to-end time to complete the job on all endpoints) as well as latency (time to complete the job on a specific endpoint). Current network traffic control algorithms do not assist with latency, as they have no application knowledge and cannot infer priority from the packet traffic coming from the management server. There is therefore a need to improve latency at both the management server and managed endpoints.